An engine model construction method includes generating test patterns in which a plurality of manipulated variables used for an engine test are changed with time, correcting the test patterns based on first coverage of a first space of manipulated variables are allowed to take and second coverage of a second space of change rate values of the manipulated variables are allowed to take, acquiring pieces of time series data of operation amounts of the manipulated variables and controlled amounts with respect to the manipulated variables by performing an engine test using the corrected test patterns, and constructing a first engine model by performing machine learning on training data in which the corrected test patterns are adopted as input and the pieces of time series data are adopted as correct answers, by a processor.
Legal claims defining the scope of protection, as filed with the USPTO.
2. The engine model construction method according to claim 1, wherein the generating includes generating, as the test patterns, Chirp signals that represent temporal changes of the manipulated variables.
3. The engine model construction method according to claim 1, wherein the constructing includes constructing, as the first engine model, a machine learning model based on one of a recurrent neural network (RNN) and a long short term memory (LSTM) with two or more intermediate layers.
4. The engine model construction method according to claim 3, further including generating a second engine model using a mathematical expression based on physics for calibrating a model parameter of the first engine model by using the first engine model.
5. The engine model construction method according to claim 1, further including generating a second engine model by linearizing the first engine model.
6. The engine model construction method according to claim 4, further including transmitting the second engine model to a second computer that predicts a controlled amount by using the first engine model.
7. The engine model construction method according to claim 1, further including at least one of excluding a region that needs to be avoided by the manipulated variables from the first space and excluding a region that needs to be avoided by the change rate values from the second space.
8. The engine model construction method according to claim 1, further including correcting the test patterns based on an air excess ratio.
10. The non-transitory computer-readable recording medium according to claim 9, wherein the generating includes generating, as the test patterns, Chirp signals that represent temporal changes of the manipulated variables.
11. The non-transitory computer-readable recording medium according to claim 9, wherein the constructing includes constructing, as the first engine model, a machine learning model based on one of a recurrent neural network (RNN) and a long short term memory (LSTM) with two or more intermediate layers.
12. The non-transitory computer-readable recording medium according to claim 11, wherein the process further includes generating a second engine model using a mathematical expression based on physics for calibrating a model parameter of the first engine model, by using the first engine model.
13. The non-transitory computer-readable recording medium according to claim 9, wherein the process further includes generating a second engine model by linearizing the first engine model.
14. The non-transitory computer-readable recording medium according to claim 12, wherein the process further includes transmitting the second engine model to a second computer that predicts a controlled amount by using the first engine model.
15. The non-transitory computer-readable recording medium according to claim 9, wherein the process further includes at least one of excluding a region that needs to be avoided by the manipulated variables from the first space and excluding a region that needs to be avoided by the change rate values from the second space.
16. The non-transitory computer-readable recording medium according to claim 9, wherein the process further includes correcting the test patterns based on an air excess ratio.
18. The engine model constructing apparatus according to claim 17, wherein the processor is further configured to generate, as the test patterns, Chirp signals that represent temporal changes of the manipulated variables.
19. The engine model constructing apparatus according to claim 17, wherein the processor is further configured to construct, as the first engine model, a machine learning model based on one of a recurrent neural network (RNN) and a long short term memory (LSTM) with two or more intermediate layers.
20. The engine model constructing apparatus according to claim 19, wherein the processor is further configured to generate a second engine model using a mathematical expression based on physics for calibrating a model parameter of the first engine model by using the first engine model.
21. The engine model constructing apparatus according to claim 17, wherein the processor is further configured to generate a second engine model by linearizing the first engine model.
22. The engine model constructing apparatus according to claim 20, wherein the processor is further configured to transmit the second engine model to a second computer that predicts a controlled amount by using the first engine model.
23. The engine model constructing apparatus according to claim 17, wherein the processor is further configured to perform at least one of excluding a region that needs to be avoided by the manipulated variables from the first space and excluding a region that needs to be avoided by the change rate values from the second space.
24. The engine model constructing apparatus according to claim 17, wherein the processor is further configured to correct the test patterns based on an air excess ratio.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
July 13, 2021
December 27, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.